llms.txt and Google Search: What It Does and Doesn't Do

Google Says llms.txt Does Nothing for Search: What That Means for Your GEO Plan

On 15 June 2026 Google's Search Central changelog settled a question that had been eating meeting time for two years. Files like llms.txt, it said, “aren't needed for Google Search (and won't negatively or positively impact your visibility or rankings)”, though maintaining them for other services is acceptable. Google's AI Optimization Guide, launched a month earlier, puts it positively: “You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search”.

Read that precisely. Google did not say llms.txt is bad, broken or penalised - it said the file is orthogonal to Google Search. No lift, no harm. Anyone who called it a ranking factor was wrong; anyone now calling the format dead is overreading the same sentence the other way.

So the honest question is not "does llms.txt work?" but "what is a zero- cost, zero- guarantee, non- Google artefact worth on a roadmap?" This article answers that, then names where the effort should go instead - the things Adviroa.ai's GEO & AEO Visibility module actually measures.

Why This Question Matters in 2026

llms.txt has taken a disproportionate share of GEO conversation because it is legible. It is a file. You can ship it in an afternoon, screenshot it, and put it on a slide as evidence the organisation takes AI search seriously. That legibility is the risk: easily demonstrated work crowds out hard- to- demonstrate work, and the hard part - crawler access, snippet eligibility, rendering - is where AI visibility is won. The June 2026 clarification is useful mostly because it removes the excuse.

What Google Actually Said, and When

Two documents, five weeks apart, saying the same thing from different angles.

• 15 June 2026, Search Central changelog: these files “aren't needed for Google Search (and won't negatively or positively impact your visibility or rankings)”. Google adds that maintaining them for other services is acceptable - this is not a discouragement.
• 15 May 2026, AI Optimization Guide: “You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search”.
• Google's AI features documentation, consistent with both: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary.”

Note the symmetry in the first quote. "Won't negatively or positively impact" rules out the optimistic reading and the anxious one at once: no penalty risk in publishing an llms.txt, and no upside in Google Search either. Genuinely neutral is a more useful answer than a warning would have been.

What llms.txt Actually Is

Most people arguing about llms.txt have not read the specification, and reading it changes the argument. llmstxt.org describes a proposal - not an adopted standard - motivated by a real constraint: “context windows are too small to handle most websites in their entirety”. The file is a curated markdown map of a site, read at inference time by an LLM that needs the right pages fast, and is explicitly not about training.

• The format, in required order: an optional BOM; an H1 with the site name - the only required section; a blockquote summary; optional detail paragraphs; then H2 sections of file lists, each entry a required markdown link of the form [name](url), optionally followed by a colon and notes.
• A section titled "Optional" has special meaning: it can be skipped when a shorter context is needed.
• Integrations exist for VitePress, Docusaurus and Drupal, so for many docs sites generation is close to free.

Read that way it is a navigation convenience, closest in spirit to a sitemap curated for a reader with a small attention budget. A reasonable idea, not a search mechanism - and its authors never claimed otherwise.

This Is Not "llms.txt Is Dead"

Google is one consumer of the open web, not all of them. Its statement covers Google Search, including AI Overviews and AI Mode, which sit downstream of ordinary Search eligibility. It says nothing about how other providers handle a curated markdown index, because Google cannot speak for them.

The accurate position is narrower than either camp wants: confirmed value of zero for Google Search, unknown vendor- dependent value elsewhere, no known downside. If your docs platform emits one automatically, there is no reason to turn it off. What you cannot do is forecast from it - no major provider has documented retrieving and privileging llms.txt, so any projected benefit is a hypothesis wearing a deliverable's clothes.

A Decision Framework for Zero- Guarantee Artefacts

llms.txt is the current example of a recurring category: cheap, plausible, unverified infrastructure aimed at a platform that never asked for it. Four questions settle each one.

1. Who is the consumer, and have they documented reading this? Without a vendor statement in writing, expected value is a guess, not a forecast.
2. What is the true cost? Not the build - ongoing accuracy. A stale index of a restructured site is worse than none.
3. What does it displace? If it takes a sprint slot that would have gone to crawler access, the file is not free at any build cost.
4. How would you know if it worked? If you cannot name the measurement in advance, ship it as maintenance, not as a project.

For llms.txt: no documented consumer, low build cost but real maintenance cost, meaningful displacement risk, no available measurement. A clear "publish if automatic, do not staff". Our companion piece on grading AI- search advice generalises this into an evidence- class framework.

Question three is the one that bites. Once llms.txt ships, "AI search" gets marked addressed and the load- bearing work loses urgency, because nothing is visibly undone. Name the reallocation in the same conversation: "we are not doing llms.txt" is weak, while "we are moving that capacity to crawler access and snippet eligibility, and here is how we will measure both" is a decision a CMO can approve.

Where to Reallocate: Four Things That Do Move Visibility

Google names its best practices plainly, and the eligibility rule for AI Overviews and AI Mode is narrow enough to audit in a day: indexed, eligible to appear in regular Search with a snippet, and meeting Search technical requirements. Four work items follow.

• Crawler access hygiene. Every major AI provider runs separate agents for training and retrieval, and blanket robots.txt blocks routinely remove sites from answer surfaces they wanted to be in - see our AI crawler access audit.
• Snippet eligibility. A nosnippet directive, or an aggressively low max- snippet, removes the page from AI Overviews and AI Mode while leaving it indexed and apparently healthy. No standard report flags it.
• Readable text. Google's stated practice is to present important content as readable text. Answers rendered only into a client- side component, or locked in an image, are not extractable whatever an auxiliary file claims.
• Internal linking. Findability through internal linking is one of Google's named practices, and it is how retrieval systems reach pages that are not your homepage.

Each is measurable, which is exactly what llms.txt is not. Adviroa.ai's GEO & AEO Visibility module runs the crawler access audit and tracks per- platform citation coverage alongside AEO and GEO scores, so the reallocation carries a number.

If You Publish One Anyway

There are good reasons to keep one: your docs platform generates it, your developer audience expects it, you want a curated entry point for internal tooling. If so, four rules.

1. Generate it from the same source of truth as your sitemap. Hand- maintained files rot within one restructure.
2. Keep it honest: pages that exist, URLs that resolve, descriptions that match.
3. Never say anything there you have not said in your HTML. If a claim lives only in llms.txt, it is not published.
4. Report it as infrastructure in the maintenance log, not as a GEO outcome in the board deck.

Adviroa.ai has an llms.txt Manager on its roadmap rather than in the product today, and we would describe it the same way if it shipped tomorrow: a hygiene convenience, not a visibility lever.

Best Practices

1. Write down which platform each GEO tactic targets. Half the confusion here comes from advice that never names its consumer.
2. Require a documented vendor statement before any tactic gets a success metric.
3. Audit crawler access, snippet directives, text rendering and internal linking before adding any file to the site root.
4. Keep auto- generated machine- readable files that cost nothing to maintain; delete stale hand- maintained ones.
5. Re- read Google's AI Optimization Guide and AI features documentation each quarter.

The Future of Machine- Readable Files for LLMs

Something here will eventually stick, because the underlying problem is real: context windows are finite and sites are large. But the version that sticks will be one a major consumer documents and commits to, arriving with a specification, a validator and a support page, the way robots.txt and sitemaps did. Until then the rational posture is cheap participation without forecasting. Watch for the first vendor to publish retrieval behaviour you can verify - that announcement, not another blog post, is the signal to change your plan.